Development of a novel clustering tool for linear peptide sequences

Abstract

Epitopes identified in large-scale screens of overlapping peptides often share significant levels of sequence identity, complicating the analysis of epitope-related data. Clustering algorithms are often used to facilitate these analyses, but available methods are generally insufficient in their capacity to define biologically meaningful epitope clusters in the context of the immune response. To fulfil this need we developed an algorithm that generates epitope clusters based on representative or consensus sequences. This tool allows the user to cluster peptide sequences on the basis of a specified level of identity by selecting among three different method options. These include the 'clique method', in which all members of the cluster must share the same minimal level of identity with each other, and the 'connected graph method', in which all members of a cluster must share a defined level of identity with at least one other member of the cluster. In cases where it is not possible to define a clear consensus sequence with the connected graph method, a third option provides a novel 'cluster-breaking algorithm' for consensus sequence driven sub-clustering. Herein we demonstrate the tool's clustering performance and applicability using (i) a selection of dengue virus epitopes for the 'clique method', (ii) sets of allergen-derived peptides from related species for the 'connected graph method' and (iii) large data sets of eluted ligand, major histocompatibility complex binding and T-cell recognition data captured within the Immune Epitope Database (IEDB) with the newly developed 'cluster-breaking algorithm'. This novel clustering tool is accessible at http://tools.iedb.org/cluster2/.

Keywords: Allergy; Antigens/Peptides/Epitopes; Bioinformatics>; MHC/HLA; Viral.

Figure 1

Plot representing the number of peptides in top 10 clusters from major histocompatibility complex class II ligand elution data.

Figure 2

Example cluster visualization before (a) and after (b) cluster‐breaking algorithms.

Figure 3

Plot representing the top 10 clusters in different data sets after applying cluster‐break algorithm.

Figure 4

Analysis of overlapping clusters in major histocompatibility complex (MHC) binding, T‐cell and MHC ligand elution data. (a) H‐chart for overlapping clusters between MHC class I binding and CD8 T‐cell assays. (b) H‐chart for overlapping clusters between MHC class II binding and CD4 T‐cell assays. (c) Pie‐chart of overlapping clusters in MHC class I ligand elution data and CD8 T‐cell assays. (d) Pie‐chart of overlapping clusters in MHC class II ligand elution data and CD4 T‐cell assays. (e) Pie‐chart of overlapping clusters in MHC class I ligand elution and binding assays data. (f) Pie‐chart of overlapping clusters in MHC class II ligand elution and binding assays data.

Figure 5

Screenshots from online tool. (a) Specify Sequence (Step 1), (b) Select clustering parameters (Step 2), (c) Choose clustering algorithm (Step 3), (d) Tabular output (Result page), (e) Graphical Output (Result page).